1

The Regulation of Unlicensed Sub-GHz bands: Are Stronger Restrictions Required for LPWAN-based IoT Success?

David Castells-Rufas, Adrià Galin-Pons, and Jordi Carrabina

 (like [4][5]) it comes implicit that a high percentage of them Abstract—Radio communications using the unlicensed Sub- will be connected by wireless links, as some of the GHz bands are expected to play an important role in the fundamental technologies enabling IoT are Low Power Wide deployment of the Internet of Things (IoT). The regulations of Area Networks (LPWAN) working on unlicensed bands, the sub-GHz unlicensed bands can affect the deployment of which are free to use. LPWAN networks in a similar way to how they affected the deployment of WLAN networks at the end of the twenty's Nevertheless, the use of the radio spectrum is regulated in century. This paper reviews the current regulations and labeling most countries of the world. This aspect is often overlooked in requirements affecting LPWAN-based IoT devices for the most the literature, not considering the limitations that regulation relevant markets worldwide (US, Europe, China, Japan, India, could impose on the deployment of such technologies. Our Brazil and Canada) and identify the main roadblocks for massive hypothesis is that current regulations can hamper the adaption of the technology. deployment of wireless IoT applications due to their impact on Finally, some suggestions are given to regulators to address the open challenges. the spectrum use and the microelectronics industries. The paper is organized as follows: we describe the radio Index Terms—Radio networks, Radio spectrum management, spectrum in Section II, and recall the events that shaped the Internet of things, Wireless sensor networks. current spectrum regulations in Section III. Section IV presents the different technologies in use in the IoT wireless I. INTRODUCTION landscape. Section V reviews the regulation and certification HERE are different predictions about the number of process on the main world markets. In Section VI we analyze Tdevices that will become connected to the internet in the what rules the regulators can enforce in trying to orchestrate near future with the widespread of the Internet of Things the spectrum. In Section VII we analyze the mathematical concept (IoT). Either being 50 G by 2020 [1], or 75 G by 2025 expressions that could describe the node density and bitrate [2], it seems to be a consensus about the disruptive nature of density of LPWAN. In Section VIII we estimate the IoT [3] and about the number of connected devices being in maximum values for LoRa and Sigfox technologies given on the order of billions. the scope of different regulations. Those results are contrasted The confluence of the evolution of many technologies like with the results from the literature in Section IX. In Section X energy scavenging, machine-to-machine communications, and we study additional economic impacts caused by the current low power wireless communication technologies support the regulation. In Section XI, before concluding, we discuss the narrative that any device that would benefit from being benefits of the harmonization of regulations. connected will definitely be connected since the cost of the connection will be insignificant. This cost includes the cost of II. RADIO-SPECTRUM LIMITS the chips, the cost of the communication channel, and the cost Although, theoretically, the radio spectrum is an infinite of the energy. resource, the interesting frequency bands for communication However, to the best of our knowledge, the studies in the over the earth surface are delimited by two factors: literature are not so explicit about predicting the number of devices that will be wirelessly connected through low power 1) In the low end, by the Shannon-Hartley theorem (Eq. 1), or low throughput radio communication links. In many works which relates the amount of information potentially transmitted over a channel.

This work has been partly funded by the Serene-IoT project (Penta 16004). D. Castells-Rufas is with the Microelectronics and Electronic Systems 퐶 = 퐵 log 1+ (1) Department, Universitat Autònoma de Barcelona Bellaterra, 08193 Spain (e- mail: [email protected]). A. Galin-Pons is with R&D Department, Applus+ Laboratories, Where 퐶 is the traffic capacity of the channel in bits per Bellaterra, 08193 Spain (e-mail: [email protected]). second, is the bandwidth of the channel in Hertz and ⁄ is Jordi Carrabina is with the Microelectronics and Electronic Systems 퐵 S 푁 Department, Universitat Autònoma de Barcelona Bellaterra, 08193 Spain (e- the signal to noise ratio of the channel. So if we want to mail: [email protected]). 2 transmit an amount of information in a time period either we For certain frequencies and the appropriate atmosphere use enough bandwidth or we have enough signal to noise ratio. conditions the ionosphere contributes to allow what is known In this trade-off we have a limited ability to increase the signal as Skywave propagation, increasing the possible range to a to noise ratio of radio channels, it is easier to select the carrier much longer distance. frequencies that will provide enough bandwidth to allow the Figure 1 depicts a simplified model of the different effects required traffic capacity. that contribute to the cost of transmitting information from a sender to a receiver using different frequencies and different 2) In the high end, frequencies above PHz are known to be distances between both endpoints. ionizing radiation and harmful to human life, so they are This is a simple model with just two endpoints. Current avoided. Secondly antenna efficiency has an intrinsic communication systems are typically not so simple, and cost is attenuation relation with frequency, i.e. a reduction of the more complex to compute. The observed limitations have been received power (푃) with respect to the emitted power (푃). overcome by deploying networks of antennas and satellite- This is known as free space path loss (FSPL). Ignoring the based communications. In this context, there is not a single gain effects of both antennas the loss is described by Eq. 2, efficiency measure, but several, like spectrum efficiency (SE) where 푑 is the distance, 푓 the frequency, and 푐 the speed of or energy efficiency (EE) [6]. light. Nevertheless, since the radio spectrum is a scarce resource, it comes as no surprise that economic laws and policymakers play an important role to orchestrate its exploitation in an 퐹푆푃퐿 = = (2) attempt to maximize its utility.

Moreover, different frequencies propagate differently in the III. BRIEF HISTORY OF RADIO SPECTRUM REGULATION atmosphere. Especially frequencies at the GHz ranges are Going back in history, after the discovery of the possibility absorbed by atmospheric gases such as O , H O, etc. 2 2 of transmitting information through electromagnetic waves, Another factor that influences the suitability of different radio was mostly used for Morse communication, but at the frequencies is the earth curvature, which limits the range of beginning with no regulation. Regulations were later direct line of sight propagation to a distance known as radio introduced in the Berlin 1903 and London 1912 conventions to horizon. The radio horizon is mainly determined by the height orchestrate different international radio services with an of the communicating antennas. important focus on emergency situations. An alternative propagation medium is the surface of the Shortly after the sinking of the Titanic, US adopted the earth. Ground wave propagation is possible below 3 MHz, but Radio Act of 1912, taking a leadership position that it would it is practically unfeasible to go further than some hundreds of maintain for the rest of the century. The main early kilometers. beneficiaries of the radio technology were still maritime ships.

The International Telecommunication Union (ITU), an International Regulatory Body (IRB) had been founded previously, in 1865. Regional and International Regulatory Bodies (RRB and IRB) were playing an important role to ensure effective communications within different territories. National Regulatory Bodies (NRB) were still not needed because the technology was either controlled by governments or in hands of very few pioneers. The invention of the amplitude modulation (AM) and its application for voice transmission caused the introduction of commercial broadcast radio stations. Soon after the first commercial radio emission by KDKA in 1920, the number of transmitters, both commercial and amateur, proliferated at a fast pace creating a chaotic situation with thousands of amateur broadcasters and common interferences to commercial radio stations. The US government saw the need of licensing different radio bands and established transmission Figure 1 Illustrative simplified example of the transmission power that a transmitter should use so that a receiver endpoint can decode the signal power limits in the Radio Act US 1927 to solve the situation. depending on the frequency of the carrier and the distance of the receiving In the following decade many advances were made. endpoint assuming an antenna height of 200m, and -120 dBm of receiver Television [7] was improved and Television broadcasters sensitivity. For lower frequencies, surface wave propagation allows a long distance range. For higher frequencies the propagation is limited to the radio appeared slowly as new users of the radio spectrum. horizon, except in the Skywave band. For extremely high frequencies above Frequency Modulation [8] was invented as a better alternative 30GHz the absorption of the wave's energy by atmospheric gases limits the to AM thanks to its lower interference features. transmissions to very short distances. In this dynamic scenario the US government issued the Communications Act of 1934, which created the Federal 3

Communications Commission (FCC), a NRB to regulate the by free market forces and less driven by the command and radio spectrum in US. Regulators not only licensed control of governments. frequencies and regulated transmission power, but also introduced the mandatory use of communication standards in IV. IOT LANDSCAPE certain licensed frequencies. For instance, in 1941 the FCC IoT is based on the idea that a myriad of devices will be created the NTSC standard making it mandatory for the VHF connected to the Internet. Some examples of these objects television channels. could be home appliances, machines, vehicles, or embedded As the regulators either limited or licensed parts of the radio sensors. Their connection will allow the acquisition of new spectrum, they raised a conflict with other uses of radio data and the opportunity to create new business models. technology that had been discovered in previous decades. In It is generally assumed that wired networks will be part of addition to telecommunications, radio could be also used for the networking infrastructure but will not provide the access induction heating, dielectric heating (microwave heating), connection to most end devices. One reason is the cost of diathermy, inducing mechanical vibration, ionization of gases, infrastructure, but another important reason is that wireless particle acceleration, etc. In order to avoid limiting the networks allow mobility. Without the need of wired advances on those technologies, the Industrial, Scientific and communication infrastructure the open challenge for wireless Medical (ISM) bands were first established at the International devices is power supply. There are four possible strategies to Telecommunications Conference of the ITU in Atlantic City power such devices: 1) connection to the power grid 2) in 1947, with the aim of allowing some unlicensed bands for rechargeable batteries 3) energy scavenging 4) life-long those applications to use free of charge. NRBs later adapted batteries. the concept introducing some limitations on emitted power The chosen strategy has a big impact on the communication and duty cycle. capabilities of such devices. Basically, as seen in Section II, Initially, it was forbidden to use unlicensed bands for the more power is available, the more bandwidth the device communications. They could exclusively be used for ISM can use. applications. But the advances in electronics and computing There are currently several available wireless technologies caused a big market pressure demanding unlicensed bands to with different properties and different target applications. allow short-range wireless communications [9]. At the same Their radio interfaces present multiple trade-offs between time, the market was also demanding permission to benefit relevant parameters which will determine the network from the advances on spread spectrum modulation, which had behavior, including: latency, mobility, cost, capacity, power been invented during the war as a military technique to consumption, complexity, reliability, interference immunity, increase the security of communication channels [10], but symmetrical uplink and downlink channels, etc. remained forbidden for civilian use. The FCC finally allowed Nevertheless, following the ETSI classification, the IoT communications on the ISM bands and the use of spread landscape can be sorted out in four main groups: spectrum in 1985. - Cellular based: all technologies based on cellular In the new scenario, regulation became very complex and a technologies optimized for IoT, including: LTE-CATM, NB- new problem arose. The risk putting a non-conformant product IoT and E-GSM. All this technologies take advantage of the in the market was high. Again, in order to protect industrial licensed band pros. investments, governments decided to mandatorily require the - Dedicated Star Networks: technologies which its pre-certification of all new products using unlicensed bands. network typology is a star and are optimized for IoT. They are The new regulations were introduced in 1989 under the FCC built over shared spectrum: Sigfox, LoRaWAN, Weightless, Part 15 rules [11]. In Europe the European Telensa, etc. Telecommunication Standard Institute (ETSI) was conceived - Dedicated Mesh Network: mesh networks covering wide the previous year in 1988. area with multi-hops connectivity -these systems are also Until our days, the following technological advances had known as Network-Based SRDs in ETSI EN 303 204-. not a big impact in the mandatory regulations of sub-GHz Silverspring technology is an example of Dedicated Mesh unlicensed bands. Nonetheless, the market pressure on the Network. continuous demand of more spectrum drove regulators to start - Low power versions of LANs & PANs: Like WiFi, mandating for higher levels of spectrum efficiency. The FCC Bluetooth (5.0/4.2/4.1/4.0, Low Energy) , WiGig, Ingenu, issued a narrowbanding mandate [12] to migrate VHF/UHF ZigBee, Thread, Z-wave, EnOcean, etc. They are also licenses to higher spectrum efficiency systems by the unlicensed technologies however the coverage range is much beginning of 2013. shorter than the second group presented above. Furthermore, from the certification perspective, private- The first two subgroups above (Cellular and dedicated star companies created different associations to promote common networks) were referred to as LPWAN by many analysts. technologies. Those associations often provide their own These two types of radio techniques share the common use of certification program, such as Wi-Fi Alliance Certification high sensitivity for increased radio coverage and the low Program, Bluetooth SIG, or Sigfox Ready to name a few. power consumption. However, there is still some discussions (like shown in The term IoT-LTN [15] refers to the Dedicated Star [13][14]) whether the spectrum use should be more controlled Networks category, which, in addition to the characteristics of 4

LPWAN, adds the properties of shared spectrum, random V. THE APPROVAL PROCESS AROUND THE GLOBE channelization, star topology and half duplex communication. The management of the radio spectrum has been assumed Table 1 presents the characteristics of the main IoT related by NRBs in most states, which implement their desired physical layers grouped by ETSI classification. In this work policies following the agreements made by RRB and IRBs. we will ignore the issues with the higher layers in the Nations must report their progress in applying the decisions International Standards Organization (ISO) communication from the ITU and the World Radiocommunication stack. Notice that dedicated star networks work in the sub- Conferences (WRC), which try to harmonize global practices. GHz bands offering a very low bitrate. They usually use As a general approach, each target market has its own modulation techniques that require less computational power regulation scheme for introducing a given RF Sub-GHz band (and energy) than the higher speed networks and can tolerate technology or generic radio transceiver as well as dedicated challenging SNRs so low as -20 dB. certification process which most heavily impacts chip The need for LTN is motivated by the type of devices that manufacturers/integrators and IoT importers, as they are are powered by life-long batteries or energy scavenging forced to spend time getting acquainted with the local legal systems. Those kind devices have a very limited energy requirements for their devices. budged that cannot be wasted on constant network connection. The process, illustrated by Figure 2, starts with the Moreover, it is well known that with modern modulations manufacturing of a system, the integration of preexisting parts receiving is more power hungry than emitting, so this limited into a system, or even with the import of a product power scenario will definitely incentivize (mostly) manufactured abroad. Each product must be tested against a unidirectional traffic from nodes to gateways that have a wired normalized test plan conceived by the regulator. The use of power supply that allows them to constantly listen to the used pre-certified modules integrated into the final host product radio channels. The need for long range coverage is the result may help to reduce the associated testing costs. In the from the economic pressure. Gateways with power supply and certification step the results of tests are analyzed together with Internet connection will usually have a much higher cost than additional technical documentation. If the process is end nodes, so it is desired to amortize their cost on the successful a label is issued, which allows the access to the maximum number of end devices. market. Some of the best candidates to take profit of such LTN In some countries local representatives are needed to be networks are Wireless Sensor Networks (WSN) [16]. able to access the market. This fact could influence the

TABLE 1 RADIO TECHNOLOGIES FOR THE PHYSICAL LAYER OF WIRELESS INTERNET OF THINGS Governing Body Frequency Capacity Category Technology Multiple Access Modulation / Standard bands (kbps) LTE-CATM 3GPP Rel 13 LTE 1024 OFDMA QPSK, 16QAM, 64QAM Cellular based NB-IoT 3GPP Rel 13 LTE/GSM 250 OFDMA BPSK, QPSK, 16QAM EG-GSM 3GPP Rel 13 GSM 240 TDMA GMSK, 8PSK Sigfox SIGFOX <1 GHz 0.6 UNB/FHSS GFSK/DBPSK Dedicated Star LoRaWAN LoRa Alliance <1 GHz 50 CSS (G)FSK Networks Weightless-P Weightless SIG <1 GHz 100 FDMA + TDMA GMSK, OQPSK Telensa WIoTF <1 GHz 0.5 UNB/FHSS 2FSK Dedicated Mesh Wi-SUN Alliance MR-FSK/MR- Silverspring <1 GHz , 2.4 GHz 1024 CSMA/CA Network IEEE 802.15.4 OFDM/MR-O-QPSK WiFi Alliance CCK, BPSK, QPSK, OFDM, DSSS, WiFi IEEE 2.4 GHz, 5 GHz 11000-6900000 16-QAM, 64-QAM, OFDMA 802.11a/b/g/n/ac 256-QAM Bluetooth Bluetooth special (4.0/4.1/4.2 interest group 2.4 GHz 1024 TDMA ASK, FSK LE) (SIG) Ingenu BPSK, OQPSK,FSK, Ingenu (formerly 20 RPMA 2.4 GHz GFSK, P-FSK, P-GFSK OnRamp) ZigBee Alliance ZigBee <1 GHz , 2.4 GHz 250 CSMA/CA DSSS, BPSK, O-QPSK Low power IEEE 802.15.4 versions of LANs Thread Group DSSS, & PANs Thread 2.4 GHz 250 CSMA/CA IEEE 802.15.4 O-QPSK Z-Wave Alliance Z-wave <1 GHz 100 TDMA FSK, GFSK ITU G.9959 EnOcean Alliance EnOcean ISO/IEC 14543- <1 GHz 125 TDMA ASK, FSK 3-1x π/2-BPSK, QPSK, WiFi Alliance WiGig 60 GHz 6760000 SC-SS QAM16, SQPSK, IEEE 802.11ad QAM64 Dash7 Dash7 Alliance <1 GHz 167 TDMA (G)FSK

5 expansion strategy of manufacturers so that they prioritize to certification step. invest in facilities where the representatives are mandatory. Additionally, the manufacturers can do the certification on One of the responsibilities of national authorities is to their own, and assume the presumption of conformity, if the perform appropriate monitoring and post-market surveillance type of device is covered by any category of the existing once the IoT-LTN devices are in the market. standards published on the Official Journal of the European Manufacturers, importers or distributors must bear in mind Union. Otherwise, the certification must be done by a Notified that, at any moment, national authorities may ask for Body. compliance exhibits. So, it is highly recommended to have Self-certification can be risky if the manufacturer is not always a product sample available. Stating that a device will versed with the European standards and the activities held by not be marketed or that is no longer manufactured is not a standardization bodies like ETSI or CENELEC. The sufficient justification for not providing post-certification applicable requirements for IoT-LTN devices fall under the production samples upon request. Short-Range Devices category regulated by ERC Recommendation 70-03 [19]. C. China

The IoT-LTN applicable standard in China is the SRRC 423 [20] (in traditional Chinese language), which list the required Figure 2 General approval procedure for new radio equipment to gain access parameters and functions that must be tested for radio to the market. transmission equipment. Testing activities shall be carried out by an Accredited Chinese Laboratory. Next, before gaining Although the process is quite similar across the world there access to the Chinese market, two certification schemes are are some differences among countries. We outline the relevant required for IoT-LTN products: an approval from the Ministry details of the process for the main global markets in terms of of Industry & Information Technology (MIIT) and the China Gross Domestic Product (GDP) (as obtained from [17]). Those Compulsory Certificate (CCC or 3C). are United States of America, the European Single Market, In addition to the typical product certification, which in englobing the European Union (EU) and European Free Trade China's case is issued by MIIT, the Chinese government Association (EFTA) states, People's Republic of China, Japan, enforces a certification on the production factories. This is Republic of India, Federative Republic of Brazil and Canada. implemented by the CCC certification that involves an audit to A. United States of America the production lines (either in China or abroad) by Chinese accredited authorities. In the United States of America, the communications The market surveillance activities are performed by the regulations are set by the FCC together with the National State Radio Monitoring and Testing Center (SRTC). Telecommunications & Information Administration (NTIA); the unlicensed equipment and intentional radiators regulations D. Japan such as Unlicensed IoT-LTN are present in the 47 CFR FCC In Japan, all the approval scheme is set by the Radio Law Rules Part 15 [18]. Testing versus those requirements shall be (Law No. 131 of May 2, 1950) which regulates the general performed by a recognized testing laboratory by the FCC. provisions for introducing a given Radio product into the The certification for equipment subject to the FCC's Japanese market, considering the applicable technical certification procedures for transmitting devices is handled by requirements, testing and certification schemes. a Telecommunication Certification Body (TCB),- a third-party Certification organizations, known as Registered organization which is devoted to review and evaluate the Certification Bodies (RCB), shall be registered by the Ministry requirements fulfilment and to upload the documentation to of Internal Affairs and Communications (MIC). MIC regulates the FCC database for approval. There are a number of TCBs the testing procedures for specified radio equipment in distributed around the globe since the FCC rules established Notification No.88 of MIC, 2004. According to the Article 38- procedures for the recognition of foreign TCBs under the 2 of the Radio Law, every type of specified radio equipment is terms of a government-to-government Mutual Recognition tested by RCBs or competent laboratories. Agreement/Arrangement (MRA). E. India B. Europe The Radio-spectrum regulations in India are driven by the In Europe, the applicable laws are derived from the Telecommunications Engineering Center (TEC), a group of Directive 2014/53/EU of the European Parliament, which the Ministry of Communications of the Indian Government. specifies the requirements on Health and Safety, The Indian Telegraph (Amendment) Rules from 2017, Electromagnetic Compatibility, and Effective use of Radio describe the test and certification scheme prior to sale, import Spectrum for new products. or use in India. There are no specific requirements on who is allowed to do The Indian Regulation consists of a collection of essential the testing step. Nevertheless, European commission names a requirements that a given device shall fulfil. Regarding the list of organizations as Notified Bodies to perform the IoT-LTN equipment, the corresponding essential requirements 6

are under TEC2449:218. All the testing activity shall be done requirements for radio apparatus that are used for radio by Indian Accredited Lab designated by TEC following the communication. Mandatory Testing and Certification of Telecom Equipments Testing laboratories test the products in accordance with the (MTCTE). enforced regulations, and certification bodies (CBs). It is Once the testing is completed and successfully possible that third party recognized independent organizations demonstrated that given device fulfils all essential certify the radio-communication equipment. requirements, the certifications must be carried out by TEC The Testing Laboratories and Certification Bodies that are Officers based on test reports and additional technical recognized by the ISED are listed on the Government of documentation. Canada website. The technical requirements for IoT-LTN devices are set on RSS-210. F. Brazil The responsible party of a given product must be within a The body taking care of the spectrum use and regulations in Canadian soil address. Foreign entities shall require a local Brazil is the Agência Nacional de Telecomunicações representative in order to start commercial activities in (ANATEL). All telecommunication products to be used in Canada. Brazil must be certified. The Regulation on The Certification and Authorization of Telecommunication Products, approved A summary of the situation in the different analyzed regions by Resolution No. 242, of 30 November 2000 establishes the is given in Table 2. general rules and procedures related to the certification and authorization of telecommunications products. VI. TECHNICAL CONSTRAINTS DERIVED FROM REGULATIONS The testing activity against the local requirements must be With spectrum management nations usually pursue the carried out by In-country test laboratory properly recognized maximization of the utility of the spectrum. As we reviewed in according to the local requirements stated by ANATEL. section III they started with a "command and control" Once the testing is carried out and given device fulfills all approach, which has been later adapted to a more market applicable technical requirements the certification takes place driven approach for certain bands [21]. It is complex to define by ANATEL. A local representative is also required according utility, but in the modern capitalist view of society, it should to the Brazilian certification scheme. have some link with a part of a nation's GDP. Following this G. Canada reasoning, regulators would aim to foster economic activity around the use of the radio spectrum (as shown in [22]). In any In Canada, it is the Innovation Science and Economic case, the job of the regulator is to select the appropriate Development Canada (ISED) that is in charge of the Radio incentives that encourage the market players to invest their Frequency Regulation and the Radio Standards Specification resources to create new wealth. (RSS). For unlicensed bands, money is not in the incentives game, The "RSS-Gen General Requirements for Compliance of so regulators select some technical parameters of radio Radio Apparatus Issue 5 (2018)" sets out the general

TABLE 2 PHY DETAILS OF THE CERTIFICATION PROCESS ACROSS THE TOP TEN GDP COUNTRIES WORLDWIDE US Europe China Japan India Brazil Canada Reference 47 CFR FCC ETSI SRRC 423 Notification TEC2449:218 Resolution No. 242 RSS-GEN Standard (test) Rules Part 15 EN 300 220-2 No.88 of MIC Resolution No. 506 RSS-247 subpart C EN 303 204 ARIB STD - §15.247 T108 Test Body Recognized Chinese Recognized Recognized Brazilian Recognized Recognized ISO 17025 Lab Own / Other ISO 17025 Lab ISO 17025 Lab ISO 17025 Lab ISO 17025 Lab ISO 17025 Lab In-country testing No No Yes No Yes Yes No required Labelling FCC ID: ISED ID: XXXXX- XXX-YYYYY YYYYY

CMIIT ID XXX - PQRS: ABCDEF XXXXX-YY-ZZZZZ 2018yznnn ABCDEF Certification Body TCB Own Producer / MIIT RCB TEC OCD CB Notify Body (DoC if HS or NB UE type examination.) Typical Lead 6 weeks 4 weeks 12 weeks 9 weeks 9 weeks 9 weeks 6 weeks Time (Test & Certificaion) National Local No No No No Yes Yes Yes Representative Required

7

transmission and decide some arbitrary thresholds following account the effects analyzed in Section II. Higher frequencies reasoned criteria. The responsibility of analyzing that all are usually used for high throughput networks (see Eq. 1). products using unlicensed bands are fulfilling the requirements Since the power required for transmission is positively is usually delegated to Certification Bodies (CBs), who verify correlated with the distance and the carrier frequency of that all the technical parameters are in the acceptable ranges. endpoints, lower frequencies are often used for longer range Although some monitoring can be done (and should be done) communications. For a given band, the regulator can establish to ensure that all the products are well behaving when a maximum Tx power limit, which almost automatically deployed in the market, it has a much lower cost for the state results in determining a maximum coverage radius (see Eq. 2). to require certification before market access. The probability of interference can be very high if no further The first decision of the regulator is to select the frequency rules are enforced, making the use of the band too bands and its applications. This is usually done taking into unpredictable for any successful business model to succeed.

TABLE 3 DETAILS OF THE TECHNICAL CONSTRAINTS BY REGULATIONS FROM THE TOP TEN GDP COUNTRIES WORLDWIDE US Europe China Japan India Brazil Canada General Parameters Frequency Range 915.9-916.9 902-907.5 902-928 863-875.6 779-787 865-867 902-928 (MHz) 920.5-929.7 915-928 Maximum TX Power 30 (>50 ch.1) 27 (869.4-869.6) 30 (>50 ch. 1) 30 (>50 ch. 1) 10 16 30 (dBm) 24 otherwise 14 (otherwise.) 24 otherwise 24 otherwise Minimum Number of 50 (BW2 < 250 kHz) 50 (BW2 < 250 kHz) 50 (BW2 < 250 kHz) - - - - Hopping Channels 25 otherwise 35 otherwise 25 otherwise Maximum Bandwidth of 500 - - - - 500 500 Hopping Channels (kHz) Maximum Spurious Emission Threshold. 54 66 66 66 66 54 54 (dBuV/m@3m) Parameters for Medium Access based on Duty Cycle Band Duty Cycle 0.1 (863-868) (%) 1 (865-868) - 0.1 (868.7-869.2) - - 1 - - 10 (869.4-869.6) 1 (870-875.6) Band Duty Cycle - 3600 - - 3600 - - Period (s) Channel Duty Cycle 2 (BW2 < 250 kHz) 2 (BW2 < 250 kHz) 2 (BW2 < 250 kHz) (%) 4 (250 kHz < BW2 < - - - - 4 (250 kHz < BW2 < 4 (250 kHz < BW2 < 500 kHz) 500Hz) 500 kHz) Channel Duty Cycle 20 (BW2 < 250 kHz) 20 (BW2 < 250 kHz) 20 (BW2 < 250 kHz) Period (s) 10 (250 Hz < BW2 < - - - - 10 (250 kHz < BW2 < 10 (250 Hz < BW2 < 500 kHz) 500 kHz) 500 kHz) Parameters for Medium Access based on Polite Spectrum Access Polite Spectrum - LBT3, AFA4 - - - - - Access Method Minimum Listening 128 (SCS5) - 160 - - - - Time Window (µs) 5000 (LCS6) Carrier Sense Level - n.a.7 - -80 - - - (dBm) 2 (SCS5 if Tx- Minimum Toff - 100 on > 6ms) - - - (ms) 50 (LCS6) Maximum 1 (single8) 0.4 (SCS5) 1 (single8) Continuous Tx-On - - - - 4 (dialoge9) 4 (LCS6) 4 (dialoge9) (s) Maximum 100s/1h over 200 100s/1h over 200 kHz of Cummulative Tx-On - kHz of the - 360s/1h (SCS5) - - the spectrum spectrum

1- Number of hopping channels 5- Short Carrier Sense 2- Bandwidth of the hopping channels 6- Long Carrier Sense 3- Listen Before Talk, a medium access method where the transmitter avoids 7- Carrier Sense Level is not defined in Europe, some indications are given in using the channel is it senses that someone is using the medium before the ETSI TR 102 313 V1.1.1 (2004-07) transmission 8- A single continuous transmission on a channel 4- Adaptive Frequency Agility, a medium access method where the transmitter 9- A multiple transmissions as part of a bidirectional protocol changes to another frequency channel if it detects that the current is being used. It can be used in conjunction with LBT.

8

So regulator usually tries to reduce that probability by VII. MAXIMUM NODE DENSITY FOR SPECTRUM USE WORST enforcing a Medium Access Policy. A possible policy is to CASE SCENARIO enforce a Band Tx Duty Cycle. That is a percentage of time Because of the expected massive deployment of IoT when the device can be actively emitting in the whole band. technology, many studies analyze the potential maximum By doing so the regulator creates the opportunity that the number of devices using a certain technology ([23][24]). channel is time multiplexed. If the endpoints would be However, those analyses are often not realistic because they perfectly coordinated, the number of potential transmitters underestimate the interference caused by other technologies would be inversely proportional to the duty cycle, but, in working on the same unlicensed bands. practice, there is no coordinator and collisions occur. As we have previously mentioned sub GHZ ISM bands are If no period is specified, there is a risk that transmitters extremely interesting for low power and long range networks take an arbitrary long cycle time as the denominator to since the required transmission power (as attenuation) has a compute the duty cycle. This would prevent others to use the quadratic relation with frequency (see Eq. 2). The low power channel for an undetermined period of time. To address this scenario generally assumes that devices will have a good issue, the regulator can specify a Band Tx Duty Cycle incentive to reduce the number of bytes transmitted to reduce Period, enforce a Maximum Band Tx-ON Time (the energy consumption, since many will run on batteries. But this maximum time that a transmitter can be actively emitting is not enforced. Nothing prevents devices connected to the continuously), or use both methods simultaneously. mains power supply from using those ISM bands. Another possible policy is to enforce a Polite Spectrum In this context, the Worst Case Scenario (WCS) analysis Access mechanism such as Listen Before Talk (LBT) or should ignore the minimum data transmission incentive and Adaptive Frequency Agility (AFA). In case of LBT the assume the maximum possible usage allowed by the regulator. regulator might specify a listening time window, and the We would like to know the maximum number of devices minimum value of the signal strength above which is transmitting on the unlicensed band in a certain area by considered signal and not noise. This value is known as assuming that they will try to work near the limits of Carrier Sense Level. Polite policies can also enforce a regulation. Since receiving is not restricted by regulation, we maximum transmission time and a Minimum Band Tx-OFF will only consider transmission. The density of transmitters Time, so that other transmitters have the chance to gain access (푛 ) will be defined by Eq. 3, where 푛 is the number of to the medium. In order to harmonize the use of the band, the regulator successful transmitters and 푎 is the area expressed in square kilometers. Thus, density of transmitters would be expressed could enforce or restrict modulation techniques or the channelization of the band, i.e., number of channels, and in devices per square km (푑푒푣/푘푚 ). channel width. 푛 = (3) On channelized bands Frequency Hoping Spread Spectrum (FHSS) can be used. If the regulator allows this, it could specify different duty cycles for each of the sub-channels, If we are using a number of channels on the frequency band while maintaining a global duty cycle for the band, or just and a duty cycle, we can observe that the total number of removing the band restriction. This is usually done by devices in a certain area is given by Eq. 4, where 푛 is the specifying a maximum transmission time on the sub-channels, simultaneous number of devices transmitting on the same which is known as Channel Dwell Time. Optionally, it is also channel, 푟 is the number of channels and ∝ is the duty cycle. possible to specify a Channel Duty Cycle, and Channel Duty Cycle Period. 푛 = (4) ∝ In such multichannel scenarios, the regulator could also decide to put a limit to the number of channels used, in other We can rewrite Eq. 3 as Eq. 5. and define 푛 as the density words, the Total Used Bandwidth. of nodes per channel. Finally, there is a need to enforce transmitters to avoid spurious emissions significant to unintended frequencies out 푛 = = = 푛 (5) of the working frequency range, which could potentially affect ∝ ∝ transmitters on licensed bands. The regulator usually specifies a Maximum spurious emission level to prevent this from Another important point in the IoT narrative is that it will happening. Table 3 summarizes some of the most important produce a huge upstream traffic of real-time data coming from values affecting the regulations for the higher frequencies of remote sensors to the Cloud. Downstream traffic is expected the unlicensed sub-GHz bands in main world markets. The to be marginal. Collected data will be stored, mined, analyzed, first observation is that the allocated frequencies are different. and visualized using BigData and (lately) Deep-Learning Other parameters also vary from country to country. In this algorithms. To understand how this goal can be achieved we context, and with the current globalization of the propose to calculate the aggregated traffic density, i.e. semiconductor industry, one can guess that this disparity of aggregated network traffic per area, which would be expressed regulations does not benefit device manufacturers. We will in bits per second per square kilometer (푏푝푠/푘푚). later insist on this issue on Section X. To compute the aggregated traffic density of the band, we 9 should sum the network traffic of all the transmitters presence of anyother transmitter will always be able to considering they are only transmitting during a duty cycle and transmit. A second transmitter will be able to transmit, only if divide them by the area, such as in Eq. 6, where 퐶 is the it is located further from the first one by a threshold distance capacity of the channel used by the transmitter 푖. d. So 푃(푇푥) = 푃(|푝 − 푝|>d). A third transmitter will be able to transmit, only if it is located further from the first and ∑ ∝ ∝ ( ) | | | | 퐶 = = = 푛 ∝ 퐶 = (6) the second. So 푃 푇푥 = 푃( 푝 − 푝 >d) 푃( 푝 − 푝 >d). Being 푝 a random variable, we can select another random variable 푤 which is the distance between two samples of p, It is interesting to realize that traffic density (퐶 ) is and we can generalize the Eq. 12 for any transmitter. independent of duty cycle. As the maximum traffic on the channel should be the total 푃(푇푥 ) = 푃(푤 > 푑) = (1− 푃(푤 ≤ 푑)) (12) band traffic capacity divided by the number of channels... Since 푃(푤 ≤ 푑) is the cumulative distribution function of 퐶 = 푟 퐶 (7) the random variable w, which can be rewritten as 퐶퐷퐹(푑), we can count how many successful transmitters there are just ...we can use (7) to rewrite (6) as (8) by adding their probabilities of success, see Eq. 13.

퐶 = (8) 푛 = ∑ 푃(푇푥) = ∑1− 퐶퐷퐹(푑)

Again it is interesting to realize that the traffic density is () 푛 = = (13) also independent of the number of radio channels in which the () band is split. So the main question again remains: what is the maximum number of simultaneous successful transmitters that But we need to know the distribution function of the distance of two random points in space. Following an analysis can coexist in a certain area using the same channel 푛? But this question is ambiguous as we should define what a similar to [25] and ignoring the corner cases we find the successful transmission is, and more important, where the expression Eq. 14. receivers of the transmissions are located, as we known (from Eq. 2) that distance is a crucial factor for the receiving power. 퐶퐷퐹 (푑) = − + (14) A. Scenario 1 Imagine that all transmitters send to a single receiver and As the area of the square tends to infinite, the density is that they are located in a radius d from it, with enough radio defined by Eq. 15. coverage. Obviously, the number of simultaneous transmissions would be 1 (푛 = 1) and the area would be the coverage circle around the receiver. In this situation, the node lim 푛 = lim = = (15) density would be defined by Eq. 10. → → 푛 = = 푛 = (10) ∝ ∝ ∝ So, this results on exactly the same expressions for 푛 as ...and the traffic density by Eq. 11 in the first scenario. In this analysis, we have used a value for the 푑 distance ∝ 퐶 = (11) equal to the coverage radius of the transmitting and receiving endpoints. This value is often empirically found depending on B. Scenario 2 the type of scenario (rural or urban), the frequency bands, and the modulation used. Imagine an infinite number of transmitters randomly located From the regulation perspective, the regulator can try to in a square of ℎ×ℎ and that all receivers are placed in a control this radius by either specifying a maximum coverage zone inside a circle of radius 푑 of its transmitter, transmission power (since limiting the transmission power such that 푑휖(0, ℎ). In this case, we should use probability limits the range) or making listen before talk mandatory and analysis to compute the maximum number of successful specifying a carrier sense level. transmitters. We denote 푇푥 as the event the transmitter i being VIII. THEORETICAL MAXIMUM DENSITIES FOR LORA AND successful and 푝 its location. We consider that any transmitter SIGFOX closer to the distance d will interfere with our signal, making it to fail. LoRa® [26] and Sigfox™ [27] are currently two popular IoT-LTN technologies. Both technologies are very different In this case, 푃(푇푥 ) =1 as the first transmitter, without the from each other and adapt to the regulatory landscape in 10 different ways. Moreover, their proposers base their business models in a different part of the value chain. TABLE 4 LoRa is promoted by Semtech Corporation, who holds LORA TYPICAL AGGREGATED CAPACITY Max Channel some patents parts on its physical channel (like [28]). It sells Bandwidth Mod. Num Capacity transceiver chips and IP to other semiconductor companies Region (kHz) Channels (bps) Total and integrators. The LoRa Alliance™ promotes the Europe 125 CSS 7 5470 38290 LoRaWAN™ networking protocol based on the LoRa 250 CSS 1 11000 11000 125 FSK 1 50000 50000 physical layer. Users can deploy their own LoRa gateways and Total 9 99290 build their network, or possibly use existing infrastructure US / CSS 125 64 5470 350080 from other organizations. There are initiatives to create Canada collaborative network infrastructure (such as The Things 500 CSS 8 12500 100000 Total 72 450080 Network [29]), but there are also traditional many telecom China 125 CSS 6 5470 32820 operators providing the infrastructure. Total 32820 On the other hand, Sigfox is promoted by the company with India 125 CSS 3 5470 16410 the same name. Sigfox also holds some patents on the physical Total 16410 layer, but their IP can be accessed freely by the members of TABLE 5 the Sigfox consortium, so IP licensing is not the core of the SIGFOX TYPICAL AGGREGATED CAPACITY business model. On the contrary, the company is focussed on Bandwidth Num Max Channel deploying the network infrastructure at the global scale and Region (kHz) Mod. Channels Capacit (bps) Total offering it as a service. Europe 0.1 D-BPSK 360 100 36000 From the technical point of view, the physical layers are US/Canada 0.6 D-BPSK 360 600 60000 very different. For this analysis we will only consider uplink TABLE 6 channels and ignore downlink ones, since this is the factor that ESTIMATED COVERAGE RADIUS FOR RURAL ENVIRONMENT ON DIFFERENT will limit the scalability of the system. LoRa uplink channels TECHNOLOGIES AND REGIONS use a Chirp Spread Spectrum (CSS) modulation, or optionally Tx Power Frequency Estimated LoRa Esimated Sigfox Frequency Shift Keying (FSK). Obviously, this requires a Region (dBm) (MHz) radius (km) radius (km) Europe 16 868 10.0 20.0 significant bandwidth, so channels use either 125 kHz or 250 US/Canada 30 915 47.5 95.0 kHz in Europe and up to 500 kHz in the US (see [30]). China 12.5 780 7.4 100.4 Sigfox uses a different approach based on Ultra Narrow India 30 866 50.2 14.8 Band (UNB) channels of 200 Hz with Binary Phase Shift Duty cycle limits must also be considered. For LoRa in US, Keying (BPSK). Canada, it would be 100% as there are no duty cycle limits Although the physical layers have a given traffic capacity, affecting the whole band. In Europe and India, the limit would the LoRaWan Alliance and Sigfox Consortium limit be 1%; and 0.1% in China. For Sigfox, the duty cycle must be themselves to a number of channels on certain frequencies to computed taking into account the daily limit of 140 messages ensure interoperability. of a maximum of 12 bytes payload per day, per device, The frequency plan of LoRa for different Regions is With all the collected information we can predict the specified in [31] . LoRa specifies some standard channels for densities for the different technologies in different regions (see every region and allows the allocation of new channels Table 7). dynamically based on applications. However, a gateway will TABLE 7 usually have a limit on the number of channels that can be DENSITY ESTIMATIONS FOR RURAL DEPLOYMENTS ON DIFFERENT listening, so to have an estimation on the typical traffic TECHNOLOGIES AND REGIONS capacity we might assume that only the standard channels are d 풏흆 푪흆 used. Technology α R (km) C (bps) (dev/km2) (bps/km2) Sigfox is using 360 channels, with a traffic capacity of 100 LoRa Europe 1% 9 10 99209 2.9 315 LoRa US/Canada 100% 72 47.5 450080 0.01 63 bps per channel in Europe and 600 bps in the US. LoRa China 0.1% 6 7.4 32820 34.9 190 Table 4 and Table 5 show the calculated aggregated traffic LoRa India 1% 3 50.2 16410 0.04 2 capacity for different regions using LoRa and Sigfox Sigfox Europe 0.0004% 360 20 36000 71619 28 technologies respectively . Sigfox US/Canada 0.0003% 360 95 60000 4232 2

Additionally, as we know from Eq. 2, the coverage radius Focusing on the network traffic capacity (see Figure 3), depends on frequency and emitting power. Since regulation is there is a big difference between different technologies and different, and we have not found an empirical analysis of their performance on various regulation landscapes. LoRa coverage in different countries, we obtain values for Obviously, there is a clear inverse relation between distance different countries applying the former equation and the and the number of bits per second that can be extracted from a maximum allowed transmission power starting with the certain area. The European version of LoRa is the technology assumption that a realistic coverage radius for a rural that offers the higher traffic capacity, but it has the drawback deployment in Europe is 10 km for LoRa and 20 km for of increasing the cost for gateway deployment. On the other Sigfox. Results are shown in Table 6. 11 extreme, the North American Sigfox offers the higher To shed some light on these issues we will have analyzed coverage but at the lowest bitrate density of less than 3 different density analysis in the literature. The different works bps/km2. In any case, the bitrate density is always below 1 are either based on analytical formulations, on simulation after kbps/km2. the characterization of the fundamental communication properties, or on the analysis of real deployments. A. Analytical studies In some analytical studies (like [24][32][33]) they analyze the deployment of an application with a certain traffic pattern. The difference with our analytical study is that they do not consider the worst case scenario imposed by regulation, but a more optimistic one. We analyze those works but only consider their reported successful transmissions. We try to harmonize the metrics so that we can compare all works.

Some works (like [24]) provide the message period (푇) between two consecutive packets from the same device, the message size in bytes (푆) , and the number of successful Figure 3 Traffic capacity density (in bits per second per square kilometer) of transmitters (푛), from which we can obtain an estimate of the LoRa and Sigfox on the analyzed regulations. Traffic capacity density axis is plotted on a logarithmic scale. total aggregate network traffic by Eq. 16.

The maximum device density (see Figure 4) is in the 퐶 = (16) expected range for Sigfox but in a lower range for LoRa, especially in US and India, where the higher coverage radius, Other works (such as [33]) provide the number of packets as a product of the higher allowed transmission power, goes per hour per node (푓 ). By a simple conversion (푇 = against the device density. Sigfox high density is the result of their self-limitation on duty cycle, but it is important to 3600⁄푓) we can find the message period and then apply understand that these numbers are ignoring the interference Eq. (16) to get network traffic as Eq. (17). between technologies. 퐶 = (17)

Table 8 shows the calculated densities derived from the information of those works, which all use LoRa technology. Coverage radius for all works is below 10 km, which seems a little optimistic in the light of many previous coverage analyses. The resulting capacity density is generally below a few hundreds of bps and the device density is only above a few hundreds of devices when low activity is assumed.

TABLE 8 ANALYTICAL STUDIES 푛 퐶 푇 푆 푛 d 퐶 (s) (B) (km) (bps) (dev/km2) (bps/km2) [32](DR5, 30 1 357 2.46 90 18.77 5 3ch. scn.1) [32](DR5, 86400 8 842710 2.46 620 44325 32 3ch.scn.2) Figure 4 Maximum device densities of Lora and Sigfox networks assuming [32](DR1, 600 20 335 7.32 80 1.99 0. 5 the maximum allowed transmission from end-devices and no interference 3ch. scn.3) from different technologies. Node density axis is plotted on a logarithmic [24](Rd. 30 1 8034 1.2 2140 1776 470 scale. Signs 6 ch.) [24] (House 86400 8 19444506 8.9 14400 78139 60 IX. STATE OF THE ART RESULTS ON DENSITIES apps. 6 ch.) [33](250 dev. 9.8 10 250 2.0 2040 19 160 Previous theoretical analysis has a limited value. First, we 3 ch.) might have been wrong in estimating the coverage radius, so [33](5K dev. 200.0 10 5000 2.0 2000 397 160 we could have been wrong on the coverage capacity and, as a 3 ch.) result, misestimated the node density for some scenarios. On B. Simulation-based studies the other hand, we have totally ignored interference, so, most probably, we have overestimated the capacity of the channel, Some analytical studies have the drawback of ignoring an overestimated the node and capacity densities. phenomena like bit error rate (BER) and interference. As 12 shown in [34][35][37], the probability of packet loss increases due to the cost of deploying a large number of devices, some as the number of nodes increases due to interference. The error recent works limit themselves to a very small number of probability has a direct relation with the time on air of the devices (like [34][35][39][40][41][42]) contributing few signal. It is known, in the case of LoRa, higher spreading interesting information rather than realistic coverage measures factors increase time on air, causing more interference errors. in different scenarios. Simulation studies usually consider scenarios with a On [43] they provide a slightly more realistic deployment number of nodes (푛) injecting an increasing number of on Congo, although with an extremely limited number of packets to the network and reporting a probability (or rate) of devices. packet transmission error (푃) or probability of packet A more important deployment is described in [44]. They transmission success (푃 =1− 푃). analyze data from the "The Things Network" over a period of For our analysis, we are going to use those probabilities (see 8 months, i.e. 21 Ms. They collected 17467312 packets with Eq. 18) to obtain the effective number of successful an average payload size of 18 bytes. This gives a total of 2.5 transmitters. Gb traffic and an effective network capacity of 119 bps. The first thing to see here is that the network is heavily underused. 푛 = 푛 푃 = 푛 1− 푃 (18) In this case, the coverage area is not reported, but they report the number of gateways 691. By a conservative 1 km2 Again, different metrics are used to report the network coverage per gateway and taking into account the reported 1618 end devices we can find a realistic value for node and traffic in the system, 푓 in [34], 푃 in [35], and 푃 in capacity densities. [36],[37],[38]. The results from another deployment in Lyon is described Most works use a quite realistic value for coverage radius of in [45] containing 10 LoRa sensors and 4 Sigfox sensors. They few km. In [35] they use a quite pessimistic value of 100m and report neither message size, nor message period. But, from the [38] uses an optimistic 6 and 12 km scenario, while [37] uses reported Daily Packet Loss statistics for LoRa sensors, we can the later. It is also strange how [36] does not locate the obtain a message frequency of 50 packets per day per sensor gateways on the center of the coverage areas. and assume a message size of 8 bytes. From the same In [38] there is no use of packet size, so we derive the information, we can derive a 푃 of 0.89. capacity by using the specified duty cycle. In [37] neither packet size, nor duty cycle is specified. In [46] the authors describe a deployment based on DQ-N, a technology based on LoRa transceivers. According to the TABLE 9 paper a DQ-N gateway supports up to 5712 nodes generating SIMULATION STUDIES an uplink traffic of 30 Bytes/hour with 36 Bytes packets. They

푇 푆 푛 d 퐶 푛 퐶 (s) (B) (km) (bps) (dev/km2) (bps/km2) do not report the exact coverage radius, but they suggest a 10 [34] (1 km typical coverage radius. In this case, no packet error 51.42 20 100 3.5 311 2.59 8 channel) probability is reported. Hence, results should be taken with a [34] (3 27.69 20 200 3.5 1156 5.19 30 grain of salt. channels) TABLE 10 [35] 1000 20 480 0.1 76 15278.87 2444 REAL DEPLOYMENTS [36] VSF 300 10 21 0.84 5 9.58 2 Naville 푇 푆 푛 d 퐶 푛 퐶 2 [36]VSF (s) (B) (km) (bps) (dev/km ) (bps/km2) 300 10 23 1.27 6 4.58 1 Saragozza [43](Congo 900 8? 13.3 0.9 0.9 5.22 0.37 [37] 400 12 0.88 Fridges) [38] 6 km 1100 6 14440 9.72 127 [44] 1618 14.83 119 2.34 0.173 [38] 12 km 600 12 7876 1.32 17 (TTN) [45] 1728 8? 10 1 0.37 2.86 0.117 [46] 4320 36 5712 10 380 18.18 1.21 The calculated densities for these works are shown in Table 9. The values for capacity density are consistently small, in a Reported densities are very low, and lower than previous similar range than the previously found with analytical simulation results. Actually the effective duty cycle of the methods. An exception is the value from [35], which is the analyzed deployments is below 0.0015%, much below the result of having a coverage radius of 100m. Such a small value regulation limits. With this parameters we can assume that not seems unacceptable for the target applications of LPWAN. much interference is happening. Regarding the density of devices, most of the values are These values are another proof that there is a need for more below 10 devices per km2, with the exception of the former research on real LTN network deployments and their case with an unacceptable coverage radius. Those values are scalability issues. significantly lower than the previously reported in analytical studies. X. THE ECONOMIC IMPACT

As detailed in [47], a typical IoT-LTN end device is an C. Real deployments. embedded device consisting of a processor connected to a Real deployments are a better source of information, but radio transceiver, a number of sensors or actuators, a power 13

supply, and the required volatile and non-volatile memory. factor between gateway and device costs is defined by X = Quite often, some of the parts can be included in the same 100X. chip. Low power embedded microcontrollers usually include As the coverage radius is increased the number of gateways the memory blocks, and there are radio transceiver SoCs needed is reduced, the aggregated cost of the needed gateways integrating most of the components. is much lower than the aggregated cost of the required IoT-LTN gateways have a significantly higher cost due to devices, and the total cost tends to the value of the devices. their higher communication needs and computing power. They However, increasing the radius reduces the node density and must receive and transmit data from several radio channels requires the duty cycle to be reduced so that all devices can be and connect with the Internet. accommodated. The cost of a deploying a 푛 number of end devices in a A. The risk for attacks region of area 푎, is given by the Eq. 19. Where X is the cost of the gateway, 푋 is the cost of an end device, ∝ is the duty The drawbacks of using an unlicensed band is that you cycle, 푟 the number of channels, and 푑 is the coverage radius cannot prevent others from using the spectrum. Anyone could of gateways. The equation is the result of assuming that the inject traffic to the air with malicious objectives. To the best gateway density is 1/푑. of our knowledge, under current regulations, this would be The 푛 devices system can only be deployed if the node totally legal. density is lower than node density value defined by Eq. 10, The most basic attack could be the jamming of the radio which basically depends on the number of channels of the channels. As seen in previous sections, coverage radius can be technology in use and the duty cycle. If the system cannot significant, and system developers must decrease the duty meet the required number of nodes, the cost is assumed to be cycle to much lower levels than those allowed by regulations infinite. to build a successful implantation. A potential attacker could jam the radio channels with a small number of end devices working at the limits of the regulation. Taking into account the 푎 + 푛푋 , < 푛 푋 = (19) low cost of the devices, the risk seems high. ∞ , otherwise As detailed in [49][50], more elaborated attacks are possible, such as the replay of emitted packets.

B. The Economics of Microelectronic Systems As seen previously, the cost for an IoT system deployment should be dominated by the end device cost, so there is a big pressure for the device producers to decrease their manufacturing cost so that they can be massively produced and deployed. The microelectronics industry is characterized by its economy of scale, so the more units of the same product you produce, the lower price you can achieve. On the other hand, different applications will need different hardware. There is a need of being able to reuse the same chips or modules for the broader possible scope. The spectrum regulation current differences among countries make it harder for industry players to meet this goal. Figure 5 Hypothetical trade-off between duty-cycle and coverage radius for Radio transceivers manufacturers have adapted to the situation the cost of the hardware needed for the deployment of 100 k devices in a 100 km2 area using LoRa with 8 channels. In this example we are assuming a cost by covering a large spectrum and allowing configuration of of 10€ for the end device and 1000€ for the gateway device. many parameters of the radio link. On the processor side, the software stack must be adapted to In this context, a system deployer should decide a system all the different behavior rules (like duty cycle requirements) with a coverage radius that minimizes the cost, but ensuring that can be easier controlled by the higher levels of the that the device density is big enough to accommodate the communication stack. expected number of devices. As detailed in [48], in the mid- A more difficult roadblock is the frequency of operation. term, the market expects a cost lower than 5$ for end-devices. The disparity on frequencies of country regulations makes the Current prices are typically higher because of the reasons we wavelength vary from 32 cm to 38 cm, or even 69 cm when will see in the following subsections. For the example we consider the 433 MHz band. This has an impact on the illustrated in Figure 5 we have assumed an end device price of antenna selection (see [51]). In order to work in all scenarios, 10€. The example shows the total cost of the hardware for the a high-bandwidth antenna should be used, but it would require deployment of 100 k devices on an area of 100 km2, with a lot of space and its price is higher than many smaller different coverage radius and duty cycle. In this example the alternatives. 14

On the other hand, lower cost antennas (like SMD ceramic, XI. A REGULATION TO FOSTER IOT-LTN APPLICATIONS or PCB based antennas) work in a much smaller bandwidth, It is clear that current regulations present a number of risks making it difficult to work on different regulations. A to the deployers of IoT systems using the unlicensed LTN mismatch between the center frequencies in such antennas can bands. This is especially problematic for network operators produce a significant efficiency reduction. In [52] they providing connectivity based on them since the uncontrolled empirically demonstrate how the coverage range of a LoRa scenario makes it extremely hard to ensure any quality of system is reduced as much as 20% when using a PCB antenna service. It is safer to invest in technologies working in licensed designed for the 915 MHz band in the 868 MHz band. bands like NB-IoT. Although the laws that limit the scalability Figure 6 illustrates different devices in the market at scale. are similar to the unlicensed case (see [53]), the market forces It is obvious that the antenna is a limiting factor for the will put adequate incentives for the correct use of the spectrum devices. A system with no antenna (a) can be as small as 12 × while preventing the most basic security attacks. However, 13 mm. A ceramic antenna (b) is one of the smallest options operators could have little incentives to deploy the followed by PCB based antenna (c). However, they have the infrastructure if very little revenue is expected, especially in drawback of the low bandwidth. On the other hand, external rural areas. In addition, the shorter range of NB-IoT (as large antennas (d) increase the cost because of the more detailed in [54]) makes it less appropriate for the rural expensive antenna, the cost of the connectors, and the bigger scenarios. Paradoxically, those scenarios are some where mechanical requirements. WSNs could benefit better from LTN connections. Some technologies like Weightless-N, already anticipate the use of licensed bands, which seems a good strategy in terms of the chip provider, but it is useless if operators do not adopt it (see [55]). Current LP-WAN technologies provide a reasonably good coverage but limit the number of potential devices when considering worst case scenarios. In the light of the findings of section IX, realistic device densities on current technologies are below 10 dev / km2. The habitable land on the earth is approximately 130 M km2. Filling the earth with IoT devices at such density factor, we would get 1.3 G devices and a total aggregated bandwidth of 130 Gbps. Even covering the whole world, this number of devices is far from the stated numbers on many optimistic forecasts. Coverage, node density, and bitrate density can be scaled- up by using directional antennas in the gateways but this has a Figure 6 a) USI's LoRa transceiver USI WM-SG-SM-42 with a connector for an external antenna. b) Miromico FMLR-72-C-STL0 fully integrated LoRa limit of just about one order of magnitude and increases the sensor with a ceramic antenna. b) LoRa transceiver module based on the cost accordingly. Reducing the coverage radius also increases Microchip's MRF89XAM8A chip and a PCB antenna. d) LoRaBee Module the cost of the infrastructure (as depicted in Figure 5) and, for RN2903 connected to an external antenna. short-range communications, there could be higher bandwidth alternatives competing with LPWAN. C. The certification overhead Regarding network traffic capacity density, all realistic The current regulation situation requires a different analyses give results below 1 kbps / km2 of unreliable traffic. certification for all countries. The management of the This automatically limits the kind of applications that can be certification processes for global deployment is an overhead, based on such traffic. but not much different than for other technologies working at Even with all the limitations, we think that unlicensed bands the 2.4 GHz band. could still be a medium for massive IoT deployment if a However, the use of different frequencies can cause to coordinated action among regulators would be adopted to create different devices for different regions. By having more minimize the risks and foster its use. than one single reference device, due to different hardware We advocate changes in the following areas: configurations, this may lead to multiple certification programs increasing the cost of testing and approval, A. A single worldwide frequency band impacting the final price of the device. The frequency band disparity among different world Trusting recognized testing laboratories and certification markets is significant. As shown in section X, it has an impact bodies is crucial for making sure that the device meets on the microelectronics industry, and the certification process. applicable regulations, confirming all related paperwork is up- A harmonization of the band would be beneficial at the to-date, and avoiding any potential market-surveillance issues global scale. The global 2.4 GHz ISM band is a good example due to non-compliances. of the benefits of such a strategy. A single band would improve the economies of scale of microelectronic chip manufacturers and allow the integration 15 of smaller antennas. It would also eliminate the need to create XII. CONCLUSION product variations to serve different markets, thereby reducing This paper has reviewed the state of the regulations and manufacturing, testing, and certification costs. Thus, the certification schemes affecting products that want to gain dimensions of end nodes would become smaller, and their cost access to the sub-GHz unlicensed band on the main world could be significantly reduced, inducing a faster adoption of markets. After an analysis of the implications of those the technology. regulations for the IoT systems, we have identified the main B. Much lower maximum duty cycles for uplink problem being the low density of end devices derived from the maximum allowed regulations parameters, such as the duty Even without considering real-world interferences, the cycle. We have identified additional problems like the risk of maximum node density with the current regulations seems too security attacks that can hinder the business models of low to justify the deployment of the infrastructure. As shown operators and system deployers. in Eq. 10, the node density is mainly determined by the duty We insist on the need to adapt and harmonize global cycle and the number of radio channels. regulations to boost the deployment of IoT so that its expected The features of Sigfox are especially adequate to allow a disruptive widespread becomes an industrial reality. Very good node density, since the UNB modulation uses a high recently, with Commission Implementing Decision (EU) number of channels and they impose themselves a very low 2018/1538 of 11 October 2018, the European Commision has duty cycle of less than 0.0004%. This results on thousands of adopted a decision to allocate a new unlicensed frequency devices per square kilometer. Nevertheless, you cannot stop band in 915-921 MHz. This is a good step towards the global others to go to much higher duty cycles, degrading the whole harmonization of unlicensed bands that could foster LPWAN- network performance, and making those numbers difficult to based IoT, achieve. In addition, the technical limitations set by regulators, and A solution would be that a low maximum duty cycle would the differences between certification schemes have also been be enforced by regulators so that high densities would be covered. The present paper also gives manufacturers, possible. The maximum duty cycle value should take into importers and general players of IoT-LTN products an account the coverage radius (see Figure 5), which is mainly overview of the current requirements for accessing some of the result of the transmitting power limit. So, the regulator the most relevant markets worldwide. should propose a combination of maximum transmitting power and maximum duty cycle to address the minimum ACKNOWLEDGMENT device density that policymakers would like to permit. This work has been partly funded by the Serene-IoT project C. Spectrum Efficiency (Penta 16004), and had inputs from 3DSafeguard (ITEA LoRa CSS and FSK channels have a spectrum efficiency of 14034), and PEM-GASOL projects. 0.04 bps/Hz and 0.4 bps/Hz while Sigfox D-BPSK has a We would like to thank Dr. Josep Parrón Granados for his spectrum efficiency of 1 bps/Hz. At first sight, UNB has a valuable comments on antenna design, Dr. Laura Prat Baiget superior spectrum efficiency than CSS, but there is still not for her help on probability functions, and Borja Herranz for enough literature about the real response with interference to his comments on commercial IoT modules. be conclusive about both technologies. Anyhow, the regulator could enforce, as it has done for other bands, modulations REFERENCES above a minimum spectrum efficiency threshold. [1] Ericsson, L. "More than 50 billion connected devices." White Paper 14 (2011): 124. D. Attacker prevention [2] Lucero, Sam. "IoT platforms: enabling the Internet of Things." White paper (2016). The proposed tougher limits on duty cycle automatically [3] Nic, N. I. C. "Disruptive civil technologies: Six technologies with would make it harder for attackers to "legally" jam the radio potential impacts on US interests out to 2025." Tech. Rep.(2008). [4] M. Centenaro, L. Vangelista, A. Zanella, and M. Zorzi, "Long-Range channels, as more devices are needed to do so. But security is Communications in Unlicensed Bands: the Rising Stars in the IoT and a serious concern. Since the risks are more complex additional Smart City Scenarios," IEEE Wireless Communications, Vol. 23, Oct. measures should be thought to improve the security of LTN 2016. [5] Biswas, Abdur Rahim, Corentin Dupont, and Congduc Pham. "IoT, networks. Cloud and BigData Integration for IoT Analytics." Building Blocks for IoT Analytics (2016): 11. E. Coordinated certification [6] G. He, S. Zhang, Y. Chen and S. Xu, "Fundamental tradeoffs and evaluation methodology for future green wireless networks," 2012 1st In order to help the industry to push the technological IEEE International Conference on Communications in China Workshops change and reduce the certification times and deployments, it (ICCC), Bejing, 2012, pp. 74-78. doi: 10.1109/ICCCW.2012.6316478 is of relevant need the setup of Mutual Recognition [7] Charles Francis Jenkins. "Transmitting pictures by wireless." U.S. Patent Agreements between different National Authorities in order to No. 1,544,156. 30 Jun. 1925 [8] Armstrong, Edwin H. "Radio broadcasting and receiving." U.S. Patent avoid extra testing and certification costs. No. 1,941,067. 26 Dec. 1933. [9] Lemstra, Wolter, Vic Hayes, and John Groenewegen. The innovation journey of Wi-Fi: The road to global success. Cambridge University Press, 2010. 16

[10] Kiesler, Markey Hedy, and Antheil George. "Secret communication [37] Georgiou, Orestis, and Usman Raza. "Low power wide area network system." U.S. Patent No. 2,292,387. 11 Aug. 1942. analysis: Can LoRa scale?." IEEE Wireless Communications Letters 6.2 [11] Carter, Kenneth R. "Unlicensed to kill: a brief history of the Part 15 (2017): 162-165. rules." info 11.5 (2009): 8-18. [38] Mahmood, Aamir, et al. "Scalability Analysis of a LoRa Network under [12] Bercovici, M. W. "FCC Narrowbanding Mandate: A Public Safety Imperfect Orthogonality." IEEE Transactions on Industrial Informatics Guide for Compliance." 2006 (2006). (2018) [13] Matinmikko, Marja, et al. "Overview and comparison of recent spectrum [39] Sanchez-Iborra, Ramon, et al. "Performance Evaluation of LoRa sharing approaches in regulation and research: From opportunistic Considering Scenario Conditions." Sensors 18.3 (2018): 772. unlicensed access towards licensed shared access." Dynamic Spectrum [40] Dvornikov, Andrey, et al. "QoS Metrics Measurement in Long Range Access Networks (DYSPAN), 2014 IEEE International Symposium on. IoT Networks." Business Informatics (CBI), 2017 IEEE 19th IEEE, 2014. Conference on. Vol. 2. IEEE, 2017. [14] Bhakthavatsalam, Anuradha C., and Arpit Khosla. "Framework for [41] Loriot, Marine, Ammar Aljer, and Isam Shahrour. "Analysis of the use demystifying M2M spectrum regulation." Australian Journal of of LoRaWan technology in a large-scale smart city demonstrator." Telecommunications and the Digital Economy 6.3 (2018): 17. Sensors Networks Smart and Emerging Technologies (SENSET), 2017. [15] Margelis, George, et al. "Low throughput networks for the IoT: Lessons IEEE, 2017. learned from industrial implementations." Internet of Things (WF-IoT), [42] Nolan, Keith E., Wael Guibene, and Mark Y. Kelly. "An evaluation of 2015 IEEE 2nd World Forum on. IEEE, 2015. low power wide area network technologies for the Internet of Things." [16] Akyildiz, Ian F., and Mehmet Can Vuran. Wireless sensor networks. Wireless Communications and Mobile Computing Conference Vol. 4. John Wiley & Sons, 2010. (IWCMC), 2016 International. IEEE, 2016. [17] IMF. World Economic Outlook, October 2018. International Monetary [43] Ramachandran, Gowri Sankar, et al. "μPnP-WAN: Experiences with Fund, 2018.. LoRa and its deployment in DR Congo." Communication Systems and [18] 47 C.F.R Chapter I – Federal Communications Commission - FCC Networks (COMSNETS), 2017 9th International Conference on. IEEE, Rules subchapter A. Available from: 2017. https://www.law.cornell.edu/cfr/text/47/chapter-I [44] Blenn, Norbert, and Fernando Kuipers. "LoRaWAN in the wild: [19] Electronic Communications Committee. "ERC Recommendation 70- Measurements from the things network." arXiv preprint 03." ECO: Tromsø, Norway (2015). arXiv:1706.03086 (2017). [20] Ministry of Industry and Information Technology of the People’s [45] Robert, Jérémy, et al. "Open IoT Ecosystem for Enhanced Republic of China. "关于发布《微功率（短距离）无线电设备的技术 Interoperability in Smart Cities—Example of Métropole De Lyon." 要求》的通知, 信部无〔2005〕423 号" Available from: Sensors 17.12 (2017): 2849. http://www.miit.gov.cn/n1146295/n1146592/n3917132/n4062354/n4062 [46] Zhang, Keyi, and Alan Marchiori. "Demo Abstract: PlanIt and DQ-N for 391/n4062397/n4062399/c4148119/part/4148120.pdf (Accessed 11 Low-Power Wide-Area Networks." Internet-of-Things Design and Nov. 2018) Implementation (IoTDI), 2017 IEEE/ACM Second International [21] Peha, Jon M. "Spectrum management policy options." IEEE Conference on. IEEE, 2017. Communications Surveys 1.1 (1998): 2-8. [47] Kruger, Carel P., and Gerhard P. Hancke. "Benchmarking Internet of [22] Yoo, Seung-Hoon, Jun-Sang Kim, and Tai-Yoo Kim. "Value-focused things devices." Industrial Informatics (INDIN), 2014 12th IEEE thinking about strategic management of radio spectrum for mobile International Conference on. IEEE, 2014 communications in Korea." Telecommunications Policy 25.10-11 [48] Boulogeorgos, Alexandros–Apostolos A., Panagiotis D. Diamantoulakis, (2001): 703-718. and George K. Karagiannidis. "Low Power Wide Area Networks [23] Lauridsen, Mads, et al. "Coverage Comparison of GPRS, NB-IoT, (LPWANs) for Internet of Things (IoT) Applications: Research LoRa, and SigFox in a 7800 km² Area." Vehicular Technology Challenges and Future Trends." arXiv preprint arXiv 1611 (2016). Conference (VTC Spring), 2017 IEEE 85th. IEEE, 2017. [49] Aras, Emekcan, et al. "Selective jamming of LoRaWAN using [24] Petäjäjärvi, Juha, et al. "Performance of a low-power wide-area network commodity hardware." arXiv preprint arXiv:1712.02141 (2017). based on LoRa technology: Doppler robustness, scalability, and [50] Aras, Emekcan, et al. "Exploring the security vulnerabilities of lora." coverage." International Journal of Distributed Sensor Networks 13.3 Cybernetics (CYBCONF), 2017 3rd IEEE International Conference on. (2017): 1550147717699412. IEEE, 2017 [25] Philip, Johan. The probability distribution of the distance between two [51] Marrocco, Gaetano. "The art of UHF RFID antenna design: Impedance- random points in a box. KTH mathematics, Royal Institute of matching and size-reduction techniques." IEEE antennas and Technology, 2007. propagation magazine 50.1 (2008). [26] Sornin, N., et al. "Lorawan specification." LoRa alliance (2015). [52] Shamanna, Pradeep. "Simple Link Budget Estimation and Performance [27] Al-Kashoash, H. A. A., and Andrew H. Kemp. "Comparison of Measurements of Microchip Sub-GHz Radio Modules." (2013) 6LoWPAN and LPWAN for the Internet of Things." Australian Journal [53] Martinez, Borja, et al. "Exploring the Performance Boundaries of NB- of Electrical and Electronics Engineering 13.4 (2016): 268-274. IoT." arXiv preprint arXiv:1810.00847 (2018). [28] Bernard, O., A. Seller, and N. Sornin. "Low power long range [54] Mekki, Kais, et al. "A comparative study of LPWAN technologies for transmitter." European Patent Application EP 2763321 (2014): A1. large-scale IoT deployment." ICT Express (2018). [29] https://www.thethingsnetwork.org/ [55] Sjöström, Daniel. "Unlicensed and licensed low-power wide area [30] Knight, Matthew, and Balint Seeber. "Decoding LoRa: Realizing a networks: Exploring the candidates for massive IoT." (2017). modern LPWAN with SDR." Proceedings of the GNU Radio Conference. Vol. 1. No. 1. 2016. [31] LoRaWAN 1.1 Regional Parameters. LoRa Alliance, Inc. (2017). [32] Mikhaylov, Konstantin, Juha Petaejaejaervi, and Tuomo Haenninen. "Analysis of capacity and scalability of the LoRa low power wide area network technology." European Wireless 2016; 22th European Wireless Conference; Proceedings of. VDE, 2016. [33] Adelantado, Ferran, et al. "Understanding the limits of LoRaWAN." IEEE Communications magazine 55.9 (2017): 34-40. [34] Haxhibeqiri, Jetmir, et al. "Lora scalability: A simulation model based on interference measurements." Sensors 17.6 (2017): 1193. [35] Bor, Martin C., et al. "Do LoRa low-power wide-area networks scale?." Proceedings of the 19th ACM International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems. ACM, 2016. [36] Pasolini, Gianni, et al. "Smart City Pilot Project Using LoRa." European Wireless 2018; 24th European Wireless Conference. VDE, 2018. 17

David Castells-Rufas was born in Jordi Carrabina was born in Manresa, Manresa, Spain, in 1971. He received the Catalonia in 1963. He graduated in B.S., M.S., and Ph.D. degrees in physics from the University Autonoma of Computer Science from Universitat Barcelona (UAB), Catalonia, Spain, in Autonoma de Barcelona (UAB), Spain in 1986, and received the M.S. and Ph.D. 1994, 2009, and 2016 respectively. degrees in Microelectronics from the From 2003 he is working as researcher Computer Science Program, UAB, in in the CEPHIS research group from the 1988 and 1991, respectively. In 1986, he Microelectronics and Electronic Systems joined the National Center for Microelectronics (CNM-CSIC), Department of the UAB, where he also teaches as a lecturer. where he was collaborating until 1996. Since 1990, he has He founded the technology based companies Histeresys and been an Associate Professor with the Department of Computer Creanium in 1998 and 2001 respectively. Science, UAB. In 2005, he joined the new Microelectronics His research interests include reconfigurable systems, high and Electronic Systems Department, heading the CEPHIS performance embedded systems, and computing architectures. research group. Since 2004, CEPHIS has been recognized as TECNIO Innovation Technology Center from the Catalan Adria Galin was born in Barcelona, Government Agency ACCIO. He is currently teaching in B.S. Spain, in 1993. He received the B.S. and M.Sc. Degree of Telecommunications Engineering and degree and M.S degree in Computer Engineering at UAB, and the Masters of Embedded Telecommunication Systems Engineering Systems at UPV-EHU and coordinating the New MsC Degree in 2015 and 2017 respectively from on IoT for eHealth. During last five years, he has coauthored Universitat Autonoma de Barcelona more than 30 papers in journals and conferences. He has been (UAB). Currently he is working as a a consultant for different international small and medium Radio-Frequency engineer in the R&D enterprises (SMEs) companies. His main interests are department of Applus+ Laboratories. He is also member of microelectronic systems oriented to embedded platforms, REDCA, TCBC, ETSI and 3GPP with active participations SoC/NoC architectures and printed microelectronics and publications in several Working Groups. He also teaches as lecturer at the Universitat Politecnica de Catalunya (UPC). Main interests include real-time instantaneous frequency estimators, testing challenges of C-V2X and 5G, regulatory compliance and interference suppression receiver techniques